Spatial and temporal analysis of rainfall and drought and development of a drought prediction model by using multi-model ensembled approach

Abstract

Drought is one of the major catastrophes faced by most of the countries in recent times. Studies have been carried out to find underlying patterns and to implement forecasting systems specified for a particular region. Finding common patterns among diverse regions and implementing forecasting systems have become challenging due to climatological differences. Climatologists have divided drought into several categories for the ease of interpretation and analysis which make analysis and forecasting more complex. Sri Lanka is a climatologically diverse country, where people can experience the climate differences within a hundred kilometres. This study contains three major components named spatial and temporal analysis, analysis of drought and development of a drought prediction model with drought risk assessment. Spatial and temporal analysis has been carried out for five selected basins in Sri Lanka namely Malwathu Oya, Kirindi Oya, Kanakarayan Aru, Gin Ganga, and Kala Oya by using selected five different drought indices. The results show that the best suited drought index to identify occurrence of drought is Standard Precipitation Index (SPI), while a significant variation is observed within Kirindi Oya basin which spans over several climatological regions. The development of the drought prediction model has been accomplished for Malwathu Oya basin by using recurrent neural networks with Long Short-Term Memory Networks and Artificial Neural Networks. The model has achieved an accuracy up to 86% in drought prediction in sub basin scale. Different models with different parameters were tested to arrive at the best suited model. A drought risk assessment has been conducted for Anuradhapura district and comparative risk in each Divisional Secretariat Division (DSD) was identified. The identified risk has been compared with the relief payments and drought affected population data in order to ensure the applicability in Anuradhapura district. The multi-model ensembled approach developed can be effectively used in drought risk identification and to obtain relative indication of socio-economic implications of drought for similar regions in Sri Lanka and elsewhere and thus can be employed as a decision support system in drought prediction and relief management.